Constrained Nonlinear Neural Model Based Predictive Control Using Genetic Algorithms

Nonlinear Model Based Predictive Control (MBPC) is one of the most powerful techniques in process control, however, two main problems need to be considered; obtaining a suitable nonlinear model and using an efficient optimization procedure. In this study, a neural network is used as a non-linear prediction model of the plant. The optimization routine is based on Genetic Algorithms (GAs). First a neural model of the non-linear system is derived from input-output data. Next, the neural model is used in an MBPC structure where the critical element is the constrained optimization routine which is no convex and thus difficult to solve. A genetic algorithm based approach is proposed to deal with this problem. The efficiency of this approach had been demonstrated with simulation examples.